Which SEC Rules Govern Big Tech's Off-Balance-Sheet Debt

Any serious analysis of big tech off-balance-sheet debt starts with a pair of numbers large enough to invite the wrong conclusion. Nikkei Asia reported in July 2026 that off-balance-sheet obligations across five U.S. tech giants had reached roughly $1.65 trillion, although the study is paywalled and the company-level methodology cannot be independently checked from the public article alone.[1] Fortune, citing Moody’s analysts David Gonzales and Alastair Drake, reported a more accounting-specific figure: $662 billion of not-yet-commenced lease commitments across the top five hyperscalers, equal to 113% of adjusted debt, and $969 billion in total undiscounted commitments.[2]

Those figures should not be translated mechanically into “hidden debt.” The legal point is narrower and more useful. These are largely footnoted obligations, disclosed through purchase-commitment, lease, consolidation, and MD&A rules. The difficult question is not whether the obligations exist somewhere outside the reader’s reach. It is whether the filer put the right obligation in the right accounting bucket, measured it consistently, described the resulting cash requirements clearly, and explained entity relationships without letting form do more work than substance.

The accounting bucket matters more than the headline number

“Off balance sheet” is a location, not a conclusion. A data-center arrangement can be a noncancelable purchase commitment, a lease that has been signed but has not commenced, an interest in a variable interest entity, or some combination of related exposures. Each classification answers a different question. A purchase commitment asks what goods, services, or capacity the company has committed to buy. A not-yet-commenced lease asks whether the company already has significant lease rights and obligations before recognition begins. A VIE analysis asks whether the company controls the economics and activities of another entity closely enough to consolidate it.

That classification discipline also explains why the numbers are hard to compare across issuers. One company’s purchase-commitment table may include short-term procurement that does not resemble another company’s long-duration data-center capacity agreement. A lease commitment measured as undiscounted future payments is not the same thing as point-in-time recognized debt. A maximum exposure disclosure for a joint venture is not a prediction that the full amount will be paid. The disclosures may be material without being interchangeable.

ASC 440: purchase commitments are affirmative footnote disclosures

ASC 440-10-50-2 is the first place to look when a hyperscaler has committed to buy capacity, equipment, services, or other inputs under a noncancelable arrangement. PwC’s summary of the rule describes disclosure for unconditional purchase obligations that are noncancelable, negotiated as part of arranging financing for facilities or equipment, and have a remaining term in excess of one year. The disclosure includes the nature and term of the obligation, the fixed or determinable amount, and aggregate amounts for each of the five succeeding fiscal years.[3]

That is not a loophole. It is a disclosure rule with gates. The filer has to decide whether the obligation is unconditional, whether it is noncancelable in the accounting sense, whether it is tied to financing arrangements of the kind contemplated by the guidance, and how the payment stream should be presented. A reader who simply totals every purchase-commitment line and calls it debt loses the distinction the accounting is trying to preserve. A filer who gives a technically compliant table but leaves the liquidity story opaque creates a different problem: the footnote may exist, yet the cash consequence may remain hard to understand.

The five-year presentation requirement is important for AI infrastructure because these arrangements often create staged obligations. The first legal question is not whether a data-center buildout is strategically exciting or alarming. It is whether the company has committed itself to future cash outflows that meet the disclosure criteria, and whether the footnote lets a reader see when those outflows are expected to occur.

ASC 842 is where the Moody’s figure becomes legally specific

The $662 billion figure reported by Fortune is analytically sharper than the broader $1.65 trillion estimate because it is tied to not-yet-commenced leases. Under ASC 842-20-50-3(b), a lessee discloses information about leases that have not yet commenced but that create significant rights and obligations for the lessee.[3] In plain terms, the contract can be important before the balance sheet catches up.

That timing point matters in data centers. A company may sign a lease or lease-like arrangement for capacity that will come online later. Before commencement, the right-of-use asset and lease liability may not yet be recognized in the same way they will be after the lease begins. But the future cash requirement can already be material, and ASC 842 gives the disclosure hook for that interim period.

This is also where investor shorthand can become legally sloppy. Undiscounted future lease commitments are not the same measurement object as adjusted debt. Comparing the two can be useful as a stress signal, which is why Moody’s 113% ratio is attention-worthy, but it should not be treated as a balance-sheet reclassification by headline. The ratio tells counsel and auditors where questions will land. It does not answer the recognition question by itself.

ASC 810 turns the inquiry from payment timing to control

The hardest cases are not always the largest purchase commitments. They are the entity-relationship cases. ASC 810 asks whether a reporting company must consolidate a variable interest entity by applying the primary-beneficiary analysis. That analysis turns on power and economics: who has the power to direct the activities that most significantly affect the entity’s economic performance, and who has exposure to benefits or losses that can be significant.

Bloomberg Tax reported that Alphabet keeps certain lease and credit-backstop variable-interest-entity arrangements off its balance sheet on the basis that it is not the primary beneficiary.[4] That is not an allegation of nondisclosure. It is a consolidation position. The disclosure risk sits in whether the analysis of power, economics, support arrangements, and exposure is adequately described and consistently applied.

Meta’s Blue Owl-related arrangement shows the same problem in a more concrete way. Bloomberg Tax reported that EY described it as “challenging” to evaluate whether Meta has power to direct the joint venture’s activities and highlighted the “significant judgment” in Meta’s consolidation assessment; the report also noted Meta’s disclosed maximum exposure of $46 billion for the venture.[4] Meta’s FY2025 Form 10-K includes related disclosures about commitments and exposures, giving readers the footnote trail that securities lawyers will later parse sentence by sentence.[5]

That is where the Enron comparison is tempting, and where it should be kept on a short leash. Special-purpose entities and variable-interest-entity mechanics invite historical memory. But the relevant question is not whether a modern data-center financing arrangement uses a structure that sounds familiar. The relevant question is the accounting mechanism: whether the reporting company is the primary beneficiary, whether it has disclosed its maximum exposure and support obligations, and whether the MD&A explains the liquidity consequences of the structure.

SEC staff are already asking about the characterization

The reason this is now a compliance issue, rather than only a footnote-reading exercise, is that SEC staff attention has moved to the same judgment calls. Bloomberg Tax reported on May 7, 2026, that Deputy Chief Accountant Sheri York said AI data-center financing is producing a “rash of accounting questions” for SEC staff and stressed “the importance of clear disclosure.”[4]

That sequencing matters. Staff are not merely reacting to large capital expenditure headlines. They are asking how arrangements are characterized, how entity relationships are analyzed, and whether disclosures make the resulting exposures understandable. Bloomberg Tax also reported that SEC Chief Accountant Kurt Hohl focused on entity relationships, which reinforces why the ASC 810 issue is not an exotic footnote to the lease discussion.[4]

For defense counsel, the staff comments create a record of foreseeability. Once the staff has publicly identified AI data-center financing as an area generating accounting questions, a later disclosure dispute will not begin from a blank page. Plaintiffs will quote the same footnote and argue that the judgment was obvious. The filer, auditor, and counsel will need to show why the judgment was reasonable at the time, based on the contract terms, accounting literature, and the total mix of disclosure.

The 2020 MD&A change made comparison harder, not disclosure optional

The SEC’s 2020 Regulation S-K modernization removed the old contractual-obligations table and replaced that standardized presentation with a principles-based requirement to discuss material cash requirements in MD&A.[6] That change did not eliminate the disclosure burden. It moved part of the burden from a grid into narrative judgment.

The old regime had its own limits. The SEC’s 2003 off-balance-sheet and aggregate contractual-obligations rules were adopted in the post-Enron period to improve visibility into obligations that might not be recognized on the balance sheet.[7] A table could be overinclusive, under-explanatory, and still difficult to compare. But it did impose a common architecture. After the 2020 amendments, companies have more room to tailor the discussion, and readers have less uniform scaffolding for cross-company comparison.

That is the current legal pressure point. A company may disclose purchase commitments under ASC 440, not-yet-commenced leases under ASC 842, VIE exposure under ASC 810, and material cash requirements in MD&A, yet still leave investors without a coherent view of how those obligations affect liquidity. Conversely, a company may provide a narrative that is clearer than a table ever was, if it explains timing, magnitude, source of obligation, and the relationship among entities.

The loss of the standardized table also complicates diligence. Counsel comparing two hyperscalers cannot safely assume that similarly named commitment lines contain similar economics. One filer may aggregate short-term procurement with longer-term infrastructure commitments. Another may emphasize lease obligations that have not commenced. A third may route the reader through a VIE exposure note and an MD&A liquidity discussion. The work is less mechanical and more dependent on tracing the obligation from contract to accounting category to narrative disclosure.

What the footnotes can and cannot prove

A footnoted commitment does not prove that management misled investors. It also does not immunize the company from a disclosure claim. The question is adequacy in context. Did the filing describe the nature of the obligation? Did it distinguish purchase commitments from leases and VIE exposures? Did it explain timing and magnitude in a way that made the cash requirement intelligible? Did it avoid burying a liquidity issue in a technical note while telling a materially different story in MD&A?

This is the bridge to litigation risk. A companion claim theory is already easy to imagine: AI growth was presented as scalable or self-funding while the company’s own footnotes showed large future infrastructure cash requirements. The legal strength of that theory would depend on the specific statements, the accounting category, the prominence of the disclosure, and what management knew when it spoke. For the litigation posture around the broader AI debt boom, see Securities Cases Emerge from the AI Debt Boom.

The practical reading is different for vendor diligence, but the same footnote discipline applies. A legal AI buyer relying on a hyperscaler’s platform should not treat every off-balance-sheet commitment as a solvency alarm. The better diligence question is whether the provider’s infrastructure model depends on commitments whose timing, renewal risk, financing support, or joint-venture structure could affect service continuity, pricing, or future strategic behavior. That is why data-center SPV disclosures, such as those discussed in Could Meta’s $14B BlackRock deal create bondholder risk?, belong in the same diligence file as security, uptime, and model-governance materials.

ASU 2023-06 is a pending wrinkle, not the current answer

There is one more rule-development point, but it should not be overstated. PwC notes that ASU 2023-06 includes commitment-disclosure changes that will not become effective unless the SEC removes related Regulation S-X or Regulation S-K provisions by June 30, 2027.[3] As of Q3 2026, that is a conditional future development, not the governing rule for current hyperscaler filings.

For now, the operative legal reading is straightforward enough to be demanding. Trace the obligation to its category. If it is a purchase commitment, test the ASC 440 criteria and the five-year disclosure. If it is a lease that has not commenced, read the ASC 842 footnote for significant rights and obligations before recognition. If a joint venture or special-purpose entity is involved, follow the ASC 810 primary-beneficiary analysis and the maximum-exposure disclosure. Then read MD&A to see whether the company made the cash requirement intelligible rather than merely locatable. SEC staff attention is not proof that the obligations are equivalent to recognized debt; it is evidence that characterization has become the compliance risk.

References

1. Five US tech giants' hidden debts soar to $1.65tn on opaque AI funding, Nikkei Asia, July 2026
2. Moody's flags $662 billion risk at the heart of the data center build-out by just 5 companies, Fortune, February 25, 2026
3. Commitments (ASC 440-10, ASC 842), PwC Viewpoint
4. SEC Calls for Clear Disclosure About AI Data Center Financing, Bloomberg Tax, May 7, 2026
5. Meta FY2025 Form 10-K, SEC EDGAR
6. SEC Amends MD&A Disclosure Rules and Trims Financial Disclosure Requirements, Proskauer
7. Disclosure in Management's Discussion and Analysis About Off-Balance Sheet Arrangements and Aggregate Contractual Obligations, SEC, 2003